Spectral Kurtosis of Choi–Williams Distribution and Hidden Markov Model for Gearbox Fault Diagnosis

Yufei Li, Wanqing Song, Fei Wu, Enrico Zio, Yujin ZhangSymmetry 2020 · DOI

TL;DR

This paper introduces a novel approach for diagnosing gearbox faults by combining the Choi-Williams distribution with spectral kurtosis (CWD-SK) for feature extraction and Hidden Markov Models (HMMs) for classification. This method effectively identifies subtle initial gearbox failures and accurately distinguishes between different fault severities, outperforming traditional methods.

What Problem Does This Paper Address?

Gearboxes are critical components in rotating machinery, and their early failure detection is crucial for preventing catastrophic damage and costly downtime. However, initial gearbox faults often manifest as faint, transient shock responses that are easily obscured by background noise, making them difficult to detect using conventional signal processing techniques. Existing methods often struggle with the non-linear and non-stationary nature of gearbox vibration signals, especially in low signal-to-noise ratio environments. The paper aims to address this challenge by developing a robust method that can accurately identify and classify gearbox faults at various stages of severity.

Methods

The proposed method integrates several key techniques:

  • Local Mean Decomposition (LMD): This algorithm is used to decompose the complex gearbox vibration signals into several product function (PF) components. This decomposition helps to isolate different signal characteristics.
  • Kurtosis Analysis: The kurtosis value of each PF component is calculated. The component with the highest kurtosis is selected, as it is more likely to contain fault-related information.
  • Choi-Williams Distribution and Spectral Kurtosis (CWD-SK): The selected PF component is then analyzed using the Choi-Williams distribution combined with spectral kurtosis. CWD-SK is a time-frequency analysis technique that is particularly adept at detecting transient signals and non-Gaussian components. A key advantage highlighted is that this approach avoids the difficult task of selecting an optimal window function, which is a common challenge with other spectral kurtosis methods.
  • Hidden Markov Models (HMMs): The extracted CWD-SK features are used as input for HMMs. HMMs are statistical models well-suited for analyzing non-stationary and sequential data, making them ideal for classifying the dynamic nature of gearbox fault signals. The paper uses HMMs for both training and diagnosis, establishing models for different fault states.

Datasets

The experimental data was collected from an MFD310 gearbox system. This system was tested under five different conditions: normal, slight fault, moderate fault, severe fault, and flaking. The vibration signals were collected with a sampling frequency of 3387.77 Hz, and 4096 sampling points were used for each signal. The paper does not introduce a new benchmark dataset but utilizes data from a real-world experimental setup.

Key Insights and Ideas

  • CWD-SK for Enhanced Transient Detection: The combination of Choi-Williams distribution and spectral kurtosis proves to be a powerful tool for highlighting transient fault signatures that are otherwise hidden in noisy vibration data.
  • Kurtosis as a Signal Quality Indicator: Using kurtosis to select the most informative PF component from LMD decomposition is an effective way to pre-process the signal and focus on the parts most likely to contain fault information.
  • HMMs for Dynamic Fault Classification: HMMs are shown to be highly effective in classifying gearbox faults due to their ability to model the temporal dependencies and non-stationary characteristics of the vibration signals.
  • Window Function Independence: A significant advantage of the CWD-SK approach is its reduced sensitivity to the choice of window function, simplifying the analysis and improving robustness.
  • Early Fault Detection Capability: The method demonstrates a clear ability to distinguish between normal and initial slight fault conditions, with the average CWD-SK value increasing notably even at the earliest stages of failure.

Key Results

  • The proposed CWD-SK combined with HMM method achieved a 95% recognition rate for classifying five gearbox states (normal, slight, moderate, severe, flaking).
  • This performance was higher than that achieved by using raw spectral kurtosis (SK) with HMM (90%) or a Back Propagation (BP) neural network with CWD-SK features (90%) and SK features (85%).
  • The CWD-SK method successfully identified initial gearbox failures, with the average CWD-SK value increasing from 1.756 in the normal state to 2.023 in the slight fault state.
  • For moderate and severe faults, the CWD-SK values became less distinguishable, highlighting the necessity of HMM for accurate classification at these stages.

Limitations and Future Work

  • Distinguishing Moderate and Severe Faults: The paper notes that the CWD-SK values for moderate and severe faults are very similar, making it difficult to differentiate them solely based on CWD-SK. This underscores the importance of the HMM for fine-grained classification.
  • Data Requirements: While the HMM model is noted to require a "small amount of data" for successful training, the exact quantity and diversity of data needed for optimal performance across various gearbox types and fault scenarios are not extensively detailed.
  • Generalizability: The study focuses on a specific MFD310 gearbox. Future work could explore the generalizability of this method to different types of gearboxes and fault mechanisms.
  • Comparison with Other Advanced Techniques: While compared to BP neural networks, further comparisons with other state-of-the-art deep learning or signal processing techniques could provide broader validation.

Summary updated

Abstract

A combination of spectral kurtosis (SK), based on Choi–Williams distribution (CWD) and hidden Markov models (HMM), accurately identifies initial gearbox failures and diagnoses fault types of gearboxes. First, using the LMD algorithm, five types of gearbox vibration signals are collected and decomposed into several product function (PF) components and the multicomponent signals are decomposed into single-component signals. Then, the kurtosis value of each component is calculated, and the component with the largest kurtosis value is selected for the CWD-SK analysis. According to the calculated CWD-SK value, the characteristics of the initial failure of the gearbox are extracted. This method not only avoids the difficulty of selecting the window function, but also provides original eigenvalues for fault feature classification. In the end, from the CWD-SK characteristic parameters at each characteristic frequency, the characteristic sequence based on CWD-SK is obtained with HMM training and diagnosis. The experimental results show that this method can effectively identify the initial fault characteristics of the gearbox, and also accurately classify the fault characteristics of different degrees.
BibTeX
@inproceedings{li2020spectral,
  author = {Yufei Li and Wanqing Song and Fei Wu and Enrico Zio and Yujin Zhang},
  title = {{Spectral Kurtosis of Choi–Williams Distribution and Hidden Markov Model for Gearbox Fault Diagnosis}},
  booktitle = {Symmetry},
  year = {2020},
  doi = {10.3390/sym12020285},
}
Figure